ShadeNet-2 20M: weights (EMA), ONNX fp32+int8, app + card
Browse files- README.md +9 -1
- assets/152029243_b3582c36fa_result.png +2 -2
- assets/154871781_ae77696b77_result.png +2 -2
- assets/157139628_5dc483e2e4_result.png +2 -2
- assets/159712188_d530dd478c_result.png +2 -2
- assets/160541986_d5be2ab4c1_result.png +2 -2
- assets/160566014_59528ff897_result.png +2 -2
- assets/160585932_fa6339f248_result.png +2 -2
- assets/160792599_6a7ec52516_result.png +2 -2
- assets/161669933_3e7d8c7e2c_result.png +2 -2
- assets/compare_152029243_b3582c36fa.png +2 -2
- assets/compare_159712188_d530dd478c.png +2 -2
- assets/compare_160585932_fa6339f248.png +2 -2
- assets/hero.png +2 -2
- checkpoints/shadenet2.ckpt +2 -2
- onnx/model.onnx +1 -1
- onnx/model_quantized.onnx +1 -1
README.md
CHANGED
|
@@ -136,10 +136,18 @@ Losses: per-map weighted MSE (depth term is **scale-shift-invariant** after leas
|
|
| 136 |
|---|---|---|---|---|---|
|
| 137 |
| last.ckpt | 0.2467 | 0.2912 | 0.0954 | 0.2541 | 0.0690 |
|
| 138 |
| best.ckpt (raw) | 0.2464 | 0.2879 | 0.0957 | 0.2561 | 0.0688 |
|
| 139 |
-
|
|
|
|
|
|
|
|
| 140 |
|
| 141 |
Full 807-image val split. Depth is the best-learned map; normals the hardest.
|
| 142 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 143 |
## Limitations
|
| 144 |
|
| 145 |
- Depth is **relative**, not metric — don't read meters off it.
|
|
|
|
| 136 |
|---|---|---|---|---|---|
|
| 137 |
| last.ckpt | 0.2467 | 0.2912 | 0.0954 | 0.2541 | 0.0690 |
|
| 138 |
| best.ckpt (raw) | 0.2464 | 0.2879 | 0.0957 | 0.2561 | 0.0688 |
|
| 139 |
+
| best.ckpt (EMA) | 0.2411 | 0.2758 | 0.0925 | 0.2542 | 0.0730 |
|
| 140 |
+
| **best-calibrated (EMA, shipped)** | **0.2384** | **0.2726** | **0.0920** | **0.2456** | 0.0787 |
|
| 141 |
+
| calibrated + 3-pass ensemble | 0.2373 | 0.2729 | 0.0917 | 0.2420 | 0.0791 |
|
| 142 |
|
| 143 |
Full 807-image val split. Depth is the best-learned map; normals the hardest.
|
| 144 |
|
| 145 |
+
Shipped weights carry a post-hoc Optuna calibration (inference-only, no
|
| 146 |
+
retraining): per-layer weight affines on the three parameter groups that a
|
| 147 |
+
sensitivity audit showed could matter (`head_out`/`head`/`up4` — everything
|
| 148 |
+
else is GroupNorm-invariant), tuned on val-loss. All page outputs use the
|
| 149 |
+
calibrated weights with a 3-pass multi-scale median ensemble.
|
| 150 |
+
|
| 151 |
## Limitations
|
| 152 |
|
| 153 |
- Depth is **relative**, not metric — don't read meters off it.
|
assets/152029243_b3582c36fa_result.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/154871781_ae77696b77_result.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/157139628_5dc483e2e4_result.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/159712188_d530dd478c_result.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/160541986_d5be2ab4c1_result.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/160566014_59528ff897_result.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/160585932_fa6339f248_result.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/160792599_6a7ec52516_result.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/161669933_3e7d8c7e2c_result.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/compare_152029243_b3582c36fa.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/compare_159712188_d530dd478c.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/compare_160585932_fa6339f248.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
assets/hero.png
CHANGED
|
Git LFS Details
|
|
Git LFS Details
|
checkpoints/shadenet2.ckpt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:debffbb12b91bad98d0b49604eaedccbdaa5f0aa44752a5800de8171204a27a3
|
| 3 |
+
size 332797787
|
onnx/model.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 80229774
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f28695e3b5e2acfafba541044a070a31fa87b6ccff4589abd8f7ba2189b24323
|
| 3 |
size 80229774
|
onnx/model_quantized.onnx
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 27964078
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a3e5a820b475b32b0cbabd7c2e1f52549d7cf24c462d3ee81125df40e0bf86e6
|
| 3 |
size 27964078
|